Responding to the geriatric ‘demographic imperative’ will represent a major challenge for emergency medicine in the coming decades. Experts have recommended a number of approaches, including structural modifications to EDs, surveillance schemes, protocols, specialized personnel, and computerized decision support to bring evidence-based diagnostics and therapies to bear upon the unique challenges in caring for these patients.50
We found that providing age-related medication computerized decision support at the point of order entry resulted in increasing the proportion of medication orders consistent with recommendations. In addition, the ADE rate was lower in the intervention group, although several secondary outcomes did not change. Findings were primarily driven by opiates, representing approximately two-thirds of all potentially inappropriate orders. Findings for other medication categories were mixed, some with a potential learning (BZDs, sedative-hypnotics) and one inter-period potential reversal effect (sedative-hypnotic).
Our study design was based loosely on a similar inpatient study conducted in our institution, which obtained somewhat different results. That study found an improvement in the prescription of the recommended total daily dose, a reduction in the incidence of 10-fold dosing orders, and in the prescription of non-recommended drugs, and was associated with a lower in-hospital fall rate, but showed no reduction in hospital length of stay. Because the inpatient study had different primary and secondary outcome measures and used a different approach for the detection of ADEs reviewed, these are not directly comparable. Because patient length of stay in the ED is measured in hours rather than days, and since the majority of orders are for a frequency of ‘×1,’ the primary outcome measures focused on the initial order for a medication rather than a total daily dose received (as was the emphasis in the inpatient study). We restricted our analysis of agreement with recommendations to initial orders for a medication, excluding subsequent orders for the same drug. Such multiple orders were infrequent, and their inclusion would have complicated analysis and made explanation cumbersome without contributing to the analysis.
Our design, consisting of alternating OFF and ON periods of computerized decision support was intended to assess whether changes in ordering behavior across study periods were sustained or rather corresponded to the active status of the decision support tool. This would indicate whether potential changes were due to real time effects of decision support versus learning that took place after the initial activation period. Since residents rotate through the ED, and likely write the majority of medication orders (though this detail was not captured), this may have been unnecessary. In order to avoid problems associated with multiple testing, we did not perform a statistical test for each inter-period interval. However, there was a significant change across periods and an apparent trend present.
The decision support tool resulted in increased agreement of medication orders with recommendations, although overall agreement with recommendations was low, and the majority of medication substitution recommendations made were declined. This is likely the result of a number of factors. The first of these is that it is notoriously difficult to change physician behavior. Although computer interventions have fared well among various strategies for impacting behavior, this phenomenon continues to be a challenging area in research and quality improvement efforts.
We believe that perhaps a more important factor may be that the medication knowledge base used in our study was not designed specifically for the ED setting. Medication and dosing recommendations that are inconsistent with an emergency physician's experience regarding efficacy and safety in the acute setting, may lead to rejection of recommended changes. While it is unclear whether or to what extent input by emergency physicians would have changed recommendations for initial drug doses, especially for this first trial in the ED, it is possible that the low to moderate compliance rates might indicate general disagreement with some of these recommendations.
Emergency patients are often of higher acuity and in higher levels of pain than those in an outpatient clinic or perhaps the medical floor, and may in particular require higher doses of analgesia than required in other settings. In addition, since emergency physicians (EPs) are physically present in the treatment area when medications are ordered and administered to patients, it may be that the feedback EPs receive on the efficacy and the dangers of commonly used medications and dosages helps avoid the more frequent pitfalls in prescribing. In this light, it is not surprising that the most common reason given for overriding recommendations for an alternate medication was that the patient had not had problems with the drug previously.
These findings are also consistent with those of a recent study of the impact of CDS on prescribing behavior for outpatient prescriptions from the ED.46
The knowledge base used in the present study shares similarities with the Beers' list, but as previously noted also includes recommendations for alternate dose, frequency, and total daily dose. It is encouraging that only 38 of the 85 unique medication/dosing recommendations were ordered in the ED. Limited applicability of existing medication recommendations for older adults to the emergency setting is not unique to our study, and has been previously discussed in the emergency medicine literature.25
It is worth noting that the development of Beers' criteria was consensus-based and that the original expert panel and subsequent updates have not included EP input. Recently, alternate criteria such as STOPP (Screening Tool of Older Persons' potentially inappropriate Prescriptions) have been proposed and studied.52
Further guideline development for use in the ED would likely benefit from site-specific adjustment.
We were interested in testing whether our intervention would result in fewer adverse events, recognizing that our study was potentially underpowered to detect this. We did find a decrease in ADEs during the periods when the CDS tool was active and that among the ADEs detected, the majority were related to medication orders not in agreement with recommendations. However, because the observed agreement rate with recommendations was low even during ON periods, this may limit the conclusions that can be drawn from this comparison.
As previously noted, because differences in ADE rates can be difficult to demonstrate, we looked at a number of secondary outcome measures intended as surrogates for adverse events, such as increased admission rate, EDLOS for discharged patients, use of rescue reversal drugs, and 10-fold drug orders. We did not find differences in these secondary outcome measures.
While the majority of recommendations from the knowledge base used in this study, in Beers' criteria, and other criteria proposed have focused on preventing PIMs or dosages thought to be associated with delirium, sedation, and other untoward outcomes, it bears consideration that there may also be risks and problems associated with under-treatment. For example, it is possible that recommendations erring toward subtherapeutic dosing of sedative-hypnotics in the setting of agitated delirium might result in an increased risk of self-harm or need for physical restraint use. Similarly, under-dosage of analgesics may result in under-treatment of pain and suffering. While the CDS knowledge base considered 5 mg of haloperidol to be a ‘10-fold order’ that might warrant review, many emergency physicians using this medication for acute agitation in the elderly might find 0.5 mg to be an ineffectively low starting dose. This reflects some of the limitations of the knowledge base which might be very appropriate for other settings or conditions but bear adjustment.
The decision-support tool and knowledge base could clearly be improved further. As noted, recommendations err on the side of conservative treatment, generally meaning alternate medications or lower dosages. Finding an appropriate range makes sense and recommendations that are sensitive to specific indications and to mental status would be ideal. It would also seem important to be able to assess the additive effects of medications of different types and among different classes. For example, if a patient were ordered a BZD and two different narcotic medications, the ordering physician would receive the same decision support as for a patient just receiving one of these three drugs. Future efforts could target such additional complexities and address errors of prescribing omission in the elderly,53
and might include some educational intervention in addition to the CDS. Most immediately, future work should include derivation of an ED-specific evidence-based knowledge base for elderly medication dosing. Decision support using this knowledge base could then be re-examined to investigate its impact on elderly ED patients.
This study has a number of limitations. The chart review portion of this study included a retrospective review. Although the previously validated screening methodology used was directed at minimizing potential hindsight bias, the inherent limitations of retrospective review with incomplete documentation cannot be completely eliminated. Emergency physicians and emergency medicine nurses are reported to miss >70% of patients with cognitive dysfunction compared to prospective data collection using validated structured screening tools.54
Chart abstraction that relies in part on documentation of delirium, for example, would potentially miss this as well. The effect of this would be to underestimate the number of ADEs manifested as confusion. We did not perform an assessment of inter-rater reliability among the nurse reviewers. However, prior internal assessments of inter-rater reliability for the same nurse reviewers using the same process has been ≥94%. Lack of follow-up on discharged ED patients who may have sustained ADEs manifested at home and who may have gone to other facilities for subsequent treatment also present limitations.
Due to technical reasons, unfortunately, orders for oxycodone/acetaminophen were not flagged as drugs in the knowledge base and so there was no decision support provided for this medication during the study periods. This was discovered too late in the study to make adjustments. Also due to technical issues in reactivating the status of the CDS after the second OFF period, the last two periods of the study were 1 week longer than the first 2 weeks. This was distributed evenly for study and control periods and we do not think it impacts the findings of the study.
The methodology in the chart review included identification of orders as PADEs if they were above the recommended dosage or frequency. Because the secondary reviews for ADEs was performed for those charts flagged as having PADEs, the likely effect of this would be to enrich the denominator of charts undergoing secondary review with those that were non-compliant, potentially biasing the results. Because of this, the significant differences observed in ADEs between compliant and non-compliant orders must be qualified accordingly, potentially over-estimating the impact of the decision support tool.
We did not collect data regarding physician demographics during the study, and thus cannot report these data. Although we collected information on reasons for declining medication substitution recommendations, we did not collect this data for when suggested dosage recommendations were declined. This was in part because this information was not included for collection in the CDS tool for the inpatient study our intervention was based on, and in part related to desires to make the CDS as least disruptive as possible. We also did not routinely collect data on patients' home medications as part of this study. Our restriction of the evaluation for ADEs to a 24 h period might conceivably limit our ability to differentiate between ADEs caused by medication ordered in the ED versus chronic medications that were previously described. Our study examined medication orders rather than medication administration. The medication administration record was not computerized and this impaired the feasibility of using this for our review. However, we do not expect this record to differ from the data obtained from our ordering data.
The study was conducted in a single institution. Because the intervention was tested at a teaching hospital where house officers write the majority of medication orders, the results may not be generalizable to other non-teaching hospital settings. We did not conduct an assessment of physician acceptability for the intervention. While the low overall compliance with the intervention may speak to acceptability of the details of the recommendations, this does not assess the process of the decision support.
Although the decision support provided includes recommendations for alternate frequency of dosing, as the large majority of medication orders in the ED have the default frequency of ‘×1’ this was not applicable. For a small number of orders frequency recommendations may have been provided. However, as the number of these was very small we did not evaluate changes in frequency as an outcome measure.
Finally, consideration of all adults ≥65 years as one population could be problematic. Confounding variables such as occult cognitive dysfunction, functional status, frailty, fall-risk, transportation deprivation, economic constraints, and insecure social safety nets may all influence the incidence of ADEs for a given medication and would ideally be considered for incorporation into future trials assessing ADE interventions in older ED populations.